About

slowbench is a notebook for things I only understood after measuring them. Most posts start the same way: I believed something about a system I built, I finally ran the numbers, and the numbers disagreed.

How the numbers here are produced

• Measurements run against the system that actually serves traffic, not a copy prepared for the benchmark.
• Gold sets, sample sizes, and the exact commands are stated in the post. If a result rests on one run, it says so.
• When a later measurement contradicts an earlier post, the earlier post gets corrected rather than quietly deleted.

What you will not find

Benchmarks I did not run, vendor comparisons I cannot reproduce, and conclusions that sounded good but were never checked.